3. Istanbul 2004
d DHM shows
les.
olation of terrain.
prediction. The
ch, however, the
will be described
of the surface is
ed distances from
puted. Each (z)
; distance value,
e surface is then
ts. A point with a
| small residual at
a low weight will
>. If an oriented
fied as off-terrain
nterpolation. This
ss are eliminated
ached. Program
for this limitation
earth surface due
this method the
led in one step,
ital. It is applied
ve setting of the
re, the base are
process yields a
m) measurement
>” of ground and
> seen as a high
vegetation points.
1e shift value for
ency is not given,
orm. This can be
| in a hierarchic
'ediction without
a. DHM
RING
for automatic
levation models
h and interpolate
be created by
AR) in an equal
spacing arrangement. The identification of points not located on the
solid ground but on topographic features like vegetation and
buildings is possible by a minimal and maximal height in the area,
by maximal height differences between neighbored points, by a
sudden change of the height level, by a linear or polynomial
interpolation in X- and Y-direction, by a minimal and maximal
height difference against a local tilted plane or polynomial surface
and a local prediction (least squares interpolation) based on the
tilted plane or polynomial surface. The final results can be filtered
(smoothened) in relation to an inclined moving plane or polynomial
surface fitted to the neighbored.
The required parameters can be automatically determined by an
analysis of the DEM based on a simple characterization of the area
as homogenous or not and smooth or undulated or mountainous.
RASCOR can respect break lines during data handling. A break
line will avoid an elimination at locations with rapid change of the
inclination like a dam.
Program RASCOR is using a sequence of different methods for
the filtering of a DSM. Only data sets with raster arrangement are
accepted.
RASCOR starts with an analysis of the height distribution itself.
This methods requires flat areas, it does not work in rolling and
mountainous terrain. It is followed by an analysis of the height
differences of neighboured points. The accepted height limit of
neighboured points is depending upon the slope and the random
errors. With this method only small objects and the boundary of
larger elements can be eliminated, but it is still very efficient.
Even large buildings can be found by a sudden change of the
elevation in a profile to a higher level and a later corresponding
change down if no vegetation is located directly beside the
buildings. This method is used for laser scanning, but it is not
optimal for DEMs determined by automatic image matching where
the buildings are looking more like hills.
Other larger objects not belonging to the bare ground are identified
by a moving local profile analysis; at first shorter and after this
longer profiles are used. The required bngth of the moving local
profiles is identified by an analysis of a sequence of shorter up to
longer profiles. In flat areas the individual height values are checked
against the mean value of the local moving profile, in rolling areas a
linear regression is used, in mountainous areas polynomials have to
be used. It will be combined with data snooping taken care about a
not even point distribution caused by previously eliminated points.
All these methods are applied in X and Y-direction. Elements
which have not been removed by this sequence of tests are
analysed by moving surfaces which may be plane, inclined or
polynomial. The size of the moving surfaces is identified by the
program itself by checking the data set with a sequence of cells
with different size. As final test a local prediction can be used, but it
is usually only finding few points not belonging to the surface after
the described sequence of tests.
In the case of the check for height differences of directly
neighboured points, the upper point will be eliminated if the
tolerance limit will be exceeded. The other methods are using a
weight factor for points located below the reference defined by the
neighboured points. This will keep points located in a ditch or
cutting in the data set. Usually points determined by laser scanning
do not have blunders causing a location below the true position,
but this may happen in the case of a DSM determined by
automatic image matching, justifying a weight factor.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
In forest areas at first only the trees are removed by the program,
smaller vegetation is remaining, so a second iteration is necessary.
A second iteration in other cases may remove also terrain points
leading to a more generalised DEM. This may be useful for the
generation of contour lines, but it $ not optimal for the correct
description of the terrain.
Highway
Highways are very often located on dams or in cuts, so
it corresponds to railways and dams. A general filtering without
taking care about the special classes tends to an elimination or
reduction of the dams, especially the upper shoulders. If only these
special segments are filtered, the filter parameters will take care
about the special structure but nevertheless the upper dam
shoulders are influenced.
standard filtering
filtering with break lines
Figure 5 highway
These elements have to be handled in a special manner. RASCOR
allows the definition of such special areas where only a local filter
will be used for the elimination of points located on cars and on
bushes on the slope. Another possibility is the use of the segment
limits and also the roads as break lines. The program will not take
care about points located behind break lines.
As it can be seen in figure 14, a standard filtering of a highway dam
is removing the trees and also bushes on the slope, but it takes also
points out of the shoulder. A standard filtering removed 21.6% of
the points. If break lines are taken into account, only 10.4% of the
points were excluded and the dam looks quite more like it should
be.